MinerU Skill
A zero-key, zero-dependency skill and CLI that turns PDF, Office and image files into Markdown via MinerU's cloud.
- 133 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The content of every processed document is sent to MinerU's cloud, and with the paid tier, together with an access token.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/mineru-skillDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The content of every processed document is sent to MinerU's cloud, and with the paid tier, together with an access token.
This is third-party code. Review the repository files before installing.
What it does
MinerU Skill wraps MinerU's cloud API in a single stdlib-only Python script. With no token and no install, the free Agent API parses a PDF, Word, PPT, Excel, HTML or image file right away and prints clean Markdown to the terminal or an agent. Setting a `MINERU_TOKEN` switches to the Standard API — files up to 200 MB and 200 pages, parallel batches, and DOCX, HTML and LaTeX export. The result can go straight into 17 delivery tools: Obsidian, Notion, Confluence, Slack, Linear, Airtable and more, via `--to` or a combination of them. A separate script exposes the same parsing as an MCP server (`mineru_mcp.py`) with the `mineru_parse`, `mineru_parse_to` and `mineru_list_sinks` tools.
Who it is for. Developers and agents who need a document turned into Markdown fast, with no account and no heavy OCR model install.
Good fit when
- An agent gets a PDF link and needs its content back as text without manually uploading the file.
- A folder of scans needs parsing and delivering straight into Obsidian or Notion in one command.
- The PDF has formulas and tables that need to survive as LaTeX and structured output, not get lost in conversion.
- A document is too big for the free tier and needs to auto-escalate to the paid API without a different command.
Not a fit when
- Documents are confidential or regulated: the tool uploads every file to MinerU's cloud, offline mode only covers born-digital PDFs.
- Maximum domain-specific accuracy for formulas and tables matters most: this is not its own recognition engine, it is a wrapper over MinerU's cloud at MinerU's accuracy.
- A RAG pipeline with structured JSON extraction and framework integration is needed: Unstructured, LlamaParse or Docling fit that better.
Example request
Parse every PDF in ./scans in parallel, skip the ones already processed, and deliver the result into my Obsidian vault.Limitations
Every file goes to MinerU's cloud; offline mode covers only born-digital PDFs via `--engine local` using pymupdf4llm. The free Agent API is capped on file size and page count; larger documents need `MINERU_TOKEN` and auto-escalate to the Standard API. Splitting oversized files (`--split`) needs the separate `pypdf` package, and RAG chunking (`--chunk`) does not replace a dedicated RAG framework. Recognition accuracy matches MinerU's cloud accuracy exactly: the tool neither improves nor degrades OCR and table quality.
How to disable. Delete the skill folder (`~/.claude/skills/mineru/` on a manual install) and unset `MINERU_TOKEN` if it was configured. Separately stop the `mineru_mcp.py` process if it was running as an MCP server.
FAQ
How do I install MinerU Skill in Claude Code or Codex?
Through Vercel Skills: `npx skills add Nebutra/MinerU-Skill` installs it straight into whichever of the 35+ supported agents you're using. The alternative is cloning manually into `~/.claude/skills/mineru/`.
Is a token required to get started?
No. The free Agent API works with no signup and no key. `MINERU_TOKEN` is only needed for files past the free size and page limits, and for DOCX, HTML or LaTeX export.
How do I deliver a result into several tools at once?
Repeat the `--to` flag: `python3 scripts/mineru.py paper.pdf --to obsidian --to notion --to slack` fans the Markdown out to all three.
How does this differ from the official MinerU MCP server?
Both wrappers give identical OCR and table output since they hit the same backend. MinerU Skill's edge is a zero-dependency core, a free token-free start, and 17 built-in delivery sinks the official MCP does not have.
Can it run offline?
Only for born-digital PDFs (text already embedded, not scans) via `--engine local`, which uses pymupdf4llm. Scans and every other format always go through the cloud.
Discussion
No questions or reviews yet.
Related
SkillsMinerU is a CLI with a document library: search and revisit an already parsed report using its local cache.
Five skills turn a research paper PDF into slides, a poster, a project page, a Xiaohongshu post and a WeChat article.
A Firecrawl skill and CLI that lets your agent convert Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and PDF to Markdown on your own machine
A skill and CLI that turn Markdown into DOCX, PDF, PPTX, XLSX, PNG and about a dozen more formats, all computed locally
An MCP server that turns PDF, DOCX, XLSX, PPTX, images, audio, web pages and YouTube transcripts into Markdown
Microsoft's utility and MCP server that turn PDF, Word, Excel, PowerPoint, HTML and other files into Markdown for models
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| MinerU Skill (this card) | no | not required | 133 | |
| MinerU | no | 83,203 | MinerU's own fully self-hosted engine; this skill is a lightweight wrapper over its cloud API with a free, key-free start | |
| MarkItDown | yes | not required | 194,189 | MarkItDown converts locally with no cloud call and without MinerU's OCR accuracy, but with no delivery into 17 tools |
| Docling | no | 70,176 | Docling parses documents fully offline; MinerU Skill trades offline capability for MinerU's cloud accuracy and ready-made delivery |
README in short
The README includes a comparison table against a dozen similar tools and names cases where a competitor fits better: offline work, RAG extraction, or above-average formula accuracy.